Abstract description generation method, abstract description model training method and computer devi

专利摘要:
This application relates to an abstract description generating method, an abstract description model training method, a computer device, and a storage medium. The abstract description generating method includes: inputting an annotated training sample into an abstract description model; performing first-phase training on an encoding network and a decoding network of the abstract description model based on supervision of a first loss function; obtaining a backward-derived hidden state of a previous moment through backward derivation according to a hidden state of each moment outputted by the decoding network; obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
公开号:EP3683725A1
申请号:EP18883654.8
申请日:2018-10-24
公开日:2020-07-22
发明作者:Xinpeng CHEN；Lin Ma；Wenhao Jiang；Wei Liu
申请人:Tencent Technology Shenzhen Co Ltd；
IPC主号:G06N3-00

专利说明:
[0001] This application claims priority to Chinese Patent Application No. 201711243949.4 , entitled "ABSTRACT DESCRIPTION GENERATING METHOD AND APPARATUS, AND ABSTRACT DESCRIPTION MODEL TRAINING METHOD AND APPARATUS" filed on November 30, 2017, which is incorporated by reference in its entirety. FIELD OF THE TECHNOLOGY
[0002] This application relates to the field of machine learning technologies, and in particular, to an abstract description generating method, an abstract description model training method, a computer device, and a storage medium. BACKGROUND OF THE DISCLOSURE
[0003] Abstract description means describing information with a statement, to obtain an abstract of the information. The information may be an image, a video, or a text. An abstract description model is a neural network model configured to obtain an abstract of information according to the inputted information.
[0004] Generally, the abstract description model includes an encoding network and a decoding network. The encoding network is configured to extract a feature from inputted data (an image or a text) to obtain an eigenvector of the inputted data. The eigenvector is inputted to a decoding nerve to obtain a hidden state of each moment, and a word of a current moment is predicted according to the hidden state, thus outputting a descriptive statement.
[0005] In a conventional abstract description model training process, the decoding network independently predicts the word of the current moment according to the hidden state of the current moment. However, context of a descriptive statement is generally correlated, and the correlation between two neighboring hidden states is not considered in a conventional abstract decoding network, resulting in low prediction accuracy of the abstract description model. SUMMARY
[0006] Embodiments of this application provide an abstract description generating method, an abstract description model training method, a computer device, and a storage medium.
[0007] An abstract description generating method, the method being implemented by a computer device and including: obtaining input information; and inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0008] An abstract description model training method, the method being implemented by a computer device and including: inputting an annotated training sample into an abstract description model; performing first-phase training on an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using a recurrent neural network (RNN) to obtain a hidden state of each current moment; obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network; obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0009] A computer device, including: a memory and a processor, the memory storing a computer program, and the computer program, when executed by the processor, causing the processor to perform the following operations: obtaining input information; and inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0010] A computer device, including: a memory and a processor, the memory storing a computer program, and the computer program, when executed by the processor, causing the processor to perform the following operations: inputting an annotated training sample into an abstract description model; training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment; obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network; obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0011] A computer-readable storage medium, the computer-readable storage medium storing a computer program, and the computer program, when executed by a processor, causing the processor to perform the following operations: obtaining input information; and inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0012] A computer-readable storage medium, the computer-readable storage medium storing a computer program, and the computer program, when executed by a processor, causing the processor to perform the following operations: inputting an annotated training sample into an abstract description model; training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment; obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network; obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0013] Details of one or more embodiments of this application are described in the accompanying drawings and descriptions below. Other features, objectives and advantages of this application are clear from the specification, the accompanying drawings, and the claims. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] To describe the technical solutions of the embodiments of this application or the related technology more clearly, the following briefly introduces the accompanying drawings required for describing the embodiments or the related technology. Apparently, the accompanying drawings in the following description show only some embodiments of the present invention, and a person of ordinary skill in the art may still derive other drawings from these accompanying drawings without creative efforts.FIG. 1 is a schematic flowchart of an abstract description model training method in an embodiment; FIG. 2 is a flowchart of steps of obtaining a backward-derived hidden state of a previous moment through backward derivation by using a reconfiguration network in an embodiment; FIG. 3 is a schematic structural diagram of a reconfiguration network in an embodiment; FIG. 4 is a flowchart of steps of first-phase training of an abstract description model in an embodiment; FIG. 5 is a schematic structural diagram of an abstract description model in an embodiment; FIG. 6 is a schematic flowchart of an abstract description model training method in another embodiment; FIG. 7 is a structural block diagram of an abstract description model training apparatus in an embodiment; FIG. 8 is a structural block diagram of an abstract description apparatus in an embodiment; and FIG. 9 is a structural block diagram of a computer device in an embodiment. DESCRIPTION OF EMBODIMENTS
[0015] To make the foregoing and other objectives, technical solutions, and advantages of this application more comprehensible, the application is further described with reference to the accompanying drawings and embodiments below. The specific embodiments described herein are used only to interpret and not to limit this application.
[0016] As shown in FIG. 1, in an embodiment, an abstract description model training method is provided, and the method is implemented by a computer device. Referring to FIG. 1, the abstract description model training method specifically includes the following operations:S102. Input an annotated training sample into an abstract description model.
[0017] The training sample is related to an actual application of a product, and the training sample may be an image, a text, or a video. The annotated training sample includes each training sample and a descriptive statement for each training sample. For example, for a given image or text segment I , the descriptive statement corresponding to the training sample is y = {y ₁, y ₂, ... , y_i }, y_i being a word forming the descriptive statement.
[0018] The abstract description model is a training object in this embodiment. A training objective is to obtain relevant parameters of the abstract description model. The abstract description model includes an encoding network and a decoding network. A feature is extracted by using the encoding network, to obtain an eigenvector of input data. Then the eigenvector is decoded by using the decoding network. At each moment in the decoding network, a hidden state of the current moment is obtained. Therefore, a word is generated from the hidden state, and a descriptive statement is obtained after several moments.
[0019] S104. Train an encoding network and a decoding network of the abstract description mode based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment.
[0020] The encoding network obtains the eigenvector of the training sample, and the encoding network may use a convolutional neural network or the RNN. The eigenvector includes a global eigenvector and a local eigenvector, the global eigenvector being a global feature representation of a training sample, and the local eigenvector being a local feature representation of a training sample.
[0021] In this embodiment, training of the abstract description model includes two phases. In the first phase, the encoding network and the decoding network of the abstract description model are used for training according to the annotated training sample. The encoding network obtains the eigenvector of the training sample, and the decoding network decodes the eigenvector to obtain a hidden state of each current moment. The decoding network generates a word according to the hidden state of each moment. A training objective of the first phase is to make the generated word corresponding to each moment as close as possible to an actual annotated word, and the word is used as the first loss function. This is a training process of maximum likelihood estimation.
[0022] A preliminary parameter of the abstract description model can be obtained after the first-phase training is performed on the abstract description model based on the first loss function. Generally, a conventional abstract description model can be obtained according to the preliminary parameter, and the conventional abstract description model can be configured to predict a text abstract of a text, an image or a video. However, there may be such a problem that in the training process, a wordy t + 1 ′
[0023] Specifically, after step S104, the method further includes:S106. Obtain a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0024] The backward-derived hidden state of the previous moment refers to a backward-derived hidden state at a moment previous to a current moment obtained by means of backward derivation based on a hidden state of the current moment outputted by the decoding network. To be specific, the backward-derived hidden state of the previous moment is obtained by derivation. A decoding process of the decoding network is to compute a hidden state of a current moment according to a hidden state of a previous moment and input of the current moment. To be specific, there is a correlation between the hidden state of the previous moment and the hidden state of the current moment, and the hidden state of the previous moment can be derived by using such a correlation.
[0025] S108. Obtain a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0026] In this embodiment, a root-mean-square error is used to measure a difference between the backward-derived hidden state ĥ _t-1 of the previous moment obtained through backward derivation and the actual hidden state h _t-1 of the previous moment outputted by the decoding network, and the difference is used as the second loss function. An objective of the second loss function is to make the difference between the backward-derived hidden state of the previous moment and the actual hidden state of the previous moment as small as possible. A mean-square error is an expected value of a second power of a difference between an estimate of a parameter and a true value of the parameter, and is denoted as MSE. The MSE is a convenient method for measuring an "average error", and can evaluate a data change degree. A smaller value of the MSE indicates higher accuracy of description of experimental data by a prediction model.
[0027] A smaller value of the second loss function indicates a smaller backward-derived hidden state of the previous moment and a smaller actual hidden state of the previous moment. The backward-derived hidden state of the previous moment is obtained by means of backward derivation from the actual hidden state of the current moment outputted by the decoding network, and the correlation between two neighboring hidden states in the decoding network is fully considered. However, in an actual prediction process, the word generated at each moment relies on the word generated at the previous moment, and thus also relies on such a correlation. Therefore, by adding second-phase training in the training process of the abstract description model, the correlation between two neighboring hidden states in the decoding network is mined to avoid the difference between training and prediction and further improve performance of the abstract description model.
[0028] S110. Obtain a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0029] In this embodiment, a second-phase training process of the abstract description model is supervised based on the second loss function. The second-phase training process is a process of adjusting, according to the correlation between two neighboring hidden states in the decoding network, a preliminary parameter of the abstract description model determined in the first phase, and a corresponding parameter is taken as a final parameter of the abstract description model in a case that the value of the second loss function reaches the preset value. Alternatively, in the second-phase training process, a parameter corresponding to a minimum value of the second loss function is used as the final parameter of the abstract description model in a case that the number of iterations is greater than a preset maximum number of iterations.
[0030] On the basis of a conventional encoding network and decoding network, a process of obtaining the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network and performing the second-phase training is added the foregoing abstract description model training method. In the second-phase training, the value of the second loss function is obtained according to the backward-derived hidden state of the previous moment and the actual hidden state of the previous moment outputted by the decoding network, and the backward-derived hidden state of the previous moment is obtained through backward derivation by using the hidden state of the current moment outputted by the decoding network. A correlation between two neighboring hidden states in the decoding network is fully considered, and such a correlation is also relied on in an actual prediction process. Therefore, the accuracy of actual prediction can be improved. In addition, the difference between the training and the prediction can be avoided, thus further improving the performance of the abstract description model.
[0031] In an embodiment, the step of obtaining the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network includes: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconfiguration network. The reconfiguration network is connected to the decoding network, and is configured to backward derive the hidden state of the previous moment corresponding to each moment by using such a characteristic that output of the decoding network relies on the context that follows. In this embodiment, the reconfiguration network uses an RNN.
[0032] FIG. 2 is a flowchart of a step of obtaining a backward-derived hidden state of a previous moment through backward derivation by using a reconfiguration network in an embodiment. As shown in FIG. 2, the step includes:S202. Input a hidden state of each current moment outputted by a decoding network and a hidden state of a previous moment in a reconfiguration network into a long short-term memory (LSTM) hidden layer of each current moment in the reconfiguration network to obtain a hidden state of the current moment in the reconfiguration network.
[0033] FIG. 3 is a schematic structural diagram of a reconfiguration network in an embodiment. The reconfiguration network includes an LSTM hidden layer and a fully connected layer FC. Output of a hidden state of each moment in the decoding network is connected to an LSTM hidden layer of a corresponding moment in the reconfiguration network, and each LSTM hidden layer is connected to one fully connected layer.
[0034] Specifically, a specific form of reconfiguration is as follows:By using a hidden state h_t of a current moment outputted by the decoding network and a hidden stateh t − 1 ′
[0035] S204. Input the hidden state of the current moment in the reconfiguration network into a fully connected layer to obtain a backward-derived hidden state of the previous moment.
[0036] In the foregoing formula,h t ′
[0037] In this embodiment, the reconfiguration network is not limited to using the LSTM hidden layer to connect two neighboring hidden states in the decoding network. The reconfiguration network may also use network structures such as a gate recurrent unit (GRU), a multilayer perceptron (MLP), a convolutional neural network (CNN), and the like to contact neighboring hidden states.
[0038] In another embodiment, the step of obtaining the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network includes: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0039] The back propagation algorithm, as the name implies, is to solve output of a neural network from input of the neural network and adjust parameters of the abstract description model. Specifically, a method for solving unconstrained problems, such as a gradient descent method, may be used to obtain a final parameter. The back propagation algorithm includes the following steps S1 to S4: S1: Perform feedforward pass computing, and obtain activation values of L₂ , L₃ ..., and an output layer L_nl by using a forward pass formula. S2: Compute a residual for each output unit i of the nl ^th layer (output layer). S3: Compute a residual of an i ^th node of an l ^th layer for each layer, where l=nl-1, nl-2, nl-2..., and 2. S4: Compute a final partial derivative value according to the residuals.
[0040] FIG. 4 is a flowchart of steps of first-phase training of an abstract description model in an embodiment. As shown in FIG. 4, a first-phase training process includes the following operations:S402. Input an annotated training sample into an encoding network, and extract a feature of training information to obtain an eigenvector of the training information.
[0041] FIG. 5 is a schematic structural diagram of an abstract description model in an embodiment. As shown in FIG. 5, the description model includes an encoding network, a decoding network, and a reconfiguration network. Output of the encoding network is connected to input of the decoding network, and output of a hidden layers of each moment in the decoding network is connected to input of the corresponding moment in the reconfiguration network.
[0042] Specifically, a global feature of the training sample extracted by the encoding network is represented as g , and 64 local features are represented as S = {S ₁, ..., S ₆₄}. In this embodiment, based on characteristics of training samples, different encoding networks may be selected to extract features.
[0043] Specifically, when an object to be described is an image, a convolutional neural network may be used as the encoding network. The convolutional neural network has good performance on an ImageNet data set. Currently, there are many convolutional neural networks that can be used as encoding networks, for example, convolutional neural networks of the Inception-X series, and convolutional neural networks of the ResNet series.
[0044] Specifically, a vector g outputted by a pooling layer of the convolutional neural network is taken as a global feature representation of the entire image, and g herein has 1536 feature dimensions. Output s of the last Inception-C module of the convolutional neural network is then taken as a local feature representation of the image, and herein s = {s ₁, ..., s ₆₄}. Each local eigenvector s_i also has 1536 dimensions. Therefore, when an image is inputted into a convolutional encoding network, a global eigenvector g ∈ ℝ 1536
[0045] When an object to be described is text data, because the text data has a strong timing characteristic, an RNN may be used to encode the text data. In this embodiment, an LSTM neural network may be used for the text data.
[0046] A to-be-inputted text sequence is denoted as I = {I ₁, ..., I_T }. Herein, I_t is a t ^th word of a current sequence, and T is a length of the text sequence. In the LSTM, a hidden state h_t may be obtained from a hidden state h _t-1 of a previous moment t - 1 and input of a current moment. A form is as follows: h t = LSTM h t − 1I t
[0047] In the encoding network, a specific expression form of the LSTM is as follows:i t f t o t g t = σ σ σ tanh T I t h t − 1
[0048] i_t is an input gate, f_t is a forget gate, o_t is an output gate, h_t is a hidden state, g_t is output of a current moment, and x_t is input of the current moment; σ is a sigmoid function; T is a transform mapping matrix; and ⊙ is an element-by-element point multiplication operator.
[0049] In this embodiment, a hidden state h_T of moment T is taken as a global representation eigenvector g of a training text, to be specific, g = h_T. The hidden state h_t generated at each moment in the LSTM is as a local representation feature of the training text, to be specific, s = {s ₁, ..., s_T } = {h ₁, ..., h_T }.
[0050] S404. Input the eigenvector into a decoding network, to obtain a hidden state of each moment.
[0051] Specifically, the decoding network is a structure for decoding, by using an RNN, the eigenvector outputted by the encoding network. The RNN may compute the hidden state h_t of the current moment according to the hidden state h _t-1 of the previous moment and the input of the current moment. Specifically, an RNN with an attention mechanism is used for decoding. A core unit of the RNN may be LSTM. A decoding form of the decoding network is as follows:i t f t o t g t = σ σ σ tanh T x t h t − 1z t
[0052] i_t is an input gate, f_t is a forget gate, o_t is an output gate, h_t is a hidden state of moment t, and σ is a sigmoid function; T is a transform mapping matrix; tanh() is an activation function; and ⊙ is an element-by-element point multiplication operator, g_t is output of a current moment, and x_t is input of the current moment. z_t is a context vector obtained by the attention mechanism. A form is as follows:z t = f att s h t − 1= ∑ i = 1 s α s i h t − 1 ∑ j = 1 sα s j h t − 1s i
[0053] In the foregoing formula, α(s_i , h _t-1) represents relevance between a local feature s = {s ₁, ... , s_T } of input data and a previous hidden state.
[0054] S406. Generate a word corresponding to a current moment according to the hidden state of each moment.
[0055] A hidden state h_t of the current moment is obtained by the decoding network at each moment, and a wordy t + 1 ′
[0056] W is a transformation matrix that maps a hidden vector to a glossary.
[0057] S408. Obtain a value of a first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment.
[0058] The abstract description model includes the encoding network and the decoding network. The encoding network obtains an eigenvector of a training sample, and the decoding network decodes the eigenvector to obtain a hidden state of each moment. A word corresponding to a current moment is generated according to the hidden state of each moment, and a value of a first loss function is obtained according to the generated word corresponding to the current moment and a current actual annotated word.
[0059] In this embodiment, a root-mean-square error is used to measure a difference between a word corresponding to a current moment generated by the decoding network and a current actual annotated word, and the difference is used as the first loss function. An objective of the first loss function is to make the difference between the word corresponding to the current moment and the current actual annotated word as small as possible.
[0060] S410. Obtain a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0061] In this embodiment, a first-phase training process of the abstract description model is supervised based on the first loss function. When the value of the first loss function reaches the preset value, a corresponding parameter is taken as the preliminary parameter of the abstract description model. Alternatively, in the first-phase training process, a parameter corresponding to a minimum value of the first loss function is used as the preliminary parameter of the abstract description model in a case that the number of iterations is greater than a preset maximum number of iterations.
[0062] An abstract description model of an embodiment is shown in FIG. 5. The description model includes an encoding network, a decoding network, and a reconfiguration network, output of the encoding network being connected to input of the decoding network, and output of a hidden layers of each moment of the decoding network being connected to input of the corresponding moment of the reconfiguration network.
[0063] FIG. 6 is a flowchart of an abstract description model training method in an embodiment. As shown in FIG. 6, the method includes the following steps: S602. Input an annotated training sample into an abstract description model. S604. Input the annotated training sample into an encoding network, and extract a feature of training information to obtain an eigenvector of the training information. S606. Input the eigenvector into a decoding network, and obtain a hidden state of each moment. S608. Generate a word corresponding to a current moment according to the hidden state of each moment. S610. Obtain a value of a first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment. S612. Obtain a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value. S614. Obtain a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0064] Specifically, this step includes: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconfiguration network, or obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0065] S616. Obtain a value of a second loss function according to the backward-derived hidden state of the previous moment and the actual hidden state of the previous moment outputted by the decoding network.
[0066] S618. Obtain a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0067] In this method, the correlation between the two neighboring hidden states in the decoding network is fully considered, and such a correlation is also relied on in an actual prediction process. Therefore, the accuracy of actual prediction can be improved. In addition, the difference between the training and the prediction can be avoided, thus further improving the performance of the abstract description model.
[0068] In an embodiment, an abstract description generating method is provided. The method is run on a server side and is implemented by a computer device on the server side. The method includes the following steps:S1. Obtain input information.
[0069] The input information is information inputted by a user through a terminal and transmitted to the server side. In this embodiment, the server may provide services such as retrieval, classification, and recommendation. The input information may be a picture or character.
[0070] S2. Input the input information into a pre-trained abstract description model, obtain an eigenvector of the input information through an encoding network of the abstract description model, and decode the eigenvector through a decoding network of the abstract description model, to generate abstract description of the input information.
[0071] Specifically, the abstract description model is obtained through training by using the abstract description model training method in the foregoing embodiments. Specifically, the abstract description model training method is described in the foregoing embodiments and details are not described herein again. FIG. 5 shows an abstract description model of an embodiment. The abstract description generating method can be configured to perform prediction on text data, image data, or a video to generate a descriptive statement. Descriptions generated for images can be used for image scenario classification, for example, automatic summarization and classification of images in a user album, and also facilitate an image retrieval service and help people with visual impairment understand the images. For text note data, the technology may be configured to describe the meaning of the text and further serve classification and mining of the text.
[0072] FIG. 7 is a schematic structural diagram of an abstract description model training apparatus in an embodiment. As shown in FIG. 7, an abstract description model training apparatus includes an input module 702, a first-phase training module 704, a backward derivation module 706, a loss value computing module 708, and a parameter determining module 710.
[0073] The input module 702 is configured to input an annotated training sample into an abstract description model.
[0074] The first-phase training module 704 is configured to train an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment.
[0075] The backward derivation module 706 is configured to obtain a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0076] The loss value computing module 708 is configured to obtain a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0077] The parameter determining module 710 is configured to obtain a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0078] On the basis of a conventional encoding network and decoding network, a process of obtaining the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network and performing second-phase training is added in the foregoing abstract description model training apparatus. In the second-phase training, the value of the second loss function is obtained according to the backward-derived hidden state of the previous moment and the actual hidden state of the previous moment outputted by the decoding network, and the backward-derived hidden state of the previous moment is obtained through backward derivation by using the actual hidden state of the current moment outputted by the decoding network. A correlation between two neighboring hidden states in the decoding network is fully considered, and such a correlation is also relied on in an actual prediction process. Therefore, the accuracy of actual prediction can be improved. In addition, the difference between the training and the prediction can be avoided, thus further improving the performance of the abstract description model.
[0079] In another embodiment, the backward derivation module is configured to obtain the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment in a reconfiguration network.
[0080] In still another embodiment, the backward derivation module includes a reconfiguration module and a connection module.
[0081] The reconfiguration module is configured to input the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into an LSTM hidden layer of the current moment of the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network.
[0082] The connection module is configured to input the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0083] In still another embodiment, the backward derivation module is configured to obtain, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0084] In an embodiment, the first-phase training module includes: an encoding module, a decoding module, a word generating module, a computing module, and a confirming module.
[0085] The encoding module is configured to input an annotated training sample into the encoding network, and extract a feature of training information to obtain an eigenvector of the training information.
[0086] The decoding module is configured to input the eigenvector into the decoding network to obtain the hidden state of each moment.
[0087] The word generating module is configured to generate a word corresponding to a current moment according to the hidden state of each moment.
[0088] The computing module is configured to obtain a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment.
[0089] The confirming module is configured to determine whether the value of the first loss function reaches a preset value, and end the first-phase training in a case that a preset value is reached.
[0090] In an embodiment, an abstract description generating apparatus is provided. As shown in FIG. 8, the abstract description generating apparatus includes an information obtaining module 802 and a prediction module 804.
[0091] The information obtaining module 802 is configured to obtain input information.
[0092] The prediction module 804 is configured to input the input information into a pre-trained abstract description model, obtain an eigenvector of the input information through an encoding network of the abstract description model, and decode the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0093] Specifically, the abstract description generating apparatus further includes structures of modules of the abstract description module training apparatus in the foregoing embodiments, and details are not described herein again.
[0094] The abstract description generating apparatus can be configured to perform prediction on text data, image data, or a video to generate a descriptive statement. Descriptions generated for images may be used for image scenario classification, for example, automatic summarization and classification of images in a user album, and also facilitate an image retrieval service and help people with visual impairment understand the images. For text note data, the technology may be configured to describe the meaning of the text and further serve classification and mining of the text.
[0095] FIG. 9 is a diagram of an internal structure of a computer device in an embodiment. As shown in FIG. 9, the computer device includes a processor, a memory, a network interface, an input device, and a display screen that are connected through a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When executed by the processor, the computer program causes the processor to implement an abstract description model training method or an abstract description generating method. The internal memory may also store a computer program, and when executed by the processor, the computer program causes the internal memory to perform the abstract description model training method or the abstract description generating method. The display screen of the computer device may be a liquid crystal display or an electronic ink display. The input device of the computer device may be a touch layer covering the display screen, or may be a key, a trackball or a touchpad disposed on a housing of the computer device, or may be an external keyboard, touchpad, or mouse, or the like.
[0096] A person skilled in the art may understand that the structure shown in FIG. 9 is only a block diagram of partial structures related to the solution of this application and does not limit the computer device on which the solution of this application is applied. The specific computer device may include more components or fewer components than those shown in the figure, or some components may be combined, or a different component deployment may be used.
[0097] In an embodiment, the abstract description model training apparatus provided by this application may be implemented in a form of a computer program, and the computer program may be run on the computer device shown in FIG. 9. The memory of the computer device can store program modules that form the abstract description model training apparatus, for example, the input module, the first-phase training module, and the backward derivation module that are shown in FIG. 7, and the information obtaining module and the prediction module that are shown in FIG. 8. The computer program including the program modules causes the processor to perform the steps in abstract description model training in embodiments of this application.
[0098] For example, the computer device shown in FIG. 9 may perform, through the input module in the abstract description model training apparatus shown in FIG. 7, the step of inputting an annotated training sample into an abstract description model. The computer device may perform, through the first-phase training module, the step of performing first-phase training on an encoding network and a decoding network of the abstract description model based on supervision of a first loss function. The computer device may obtain, through the backward derivation module, a backward-derived hidden state of a previous moment through backward derivation according to a hidden state of each moment outputted by the decoding network.
[0099] For example again, the computer device in FIG. 9 may perform, through the input module in the abstract description generating apparatus shown in FIG. 8, the step of obtaining input information and perform, through the prediction module, the step of inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through the encoding network of the abstract description model, and decoding the eigenvector through the decoding network of the abstract description model to generate abstract description of the input information.
[0100] In an embodiment, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program. The computer program, when executed by the processor, causes the processor to perform the following operations: obtaining input information; and inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0101] In an embodiment, when executed by the processor, the computer program causes the processor to perform the following operations: inputting an annotated training sample into the abstract description model; training the encoding network and the decoding network of the abstract description model based on supervision of the first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment; obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network; obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0102] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining the backward-derived hidden state of the previous moment by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconfiguration network.
[0103] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into an LSTM hidden layer of the current moment of the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network; and inputting the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0104] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0105] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting an annotated training sample into the encoding network, and extracting a feature of training information to obtain an eigenvector of the training information; inputting the eigenvector into the decoding network to obtain a hidden state of each moment; generating a word corresponding to a current moment according to the hidden state of each moment; obtaining a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment; and obtaining a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0106] In an embodiment, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program. The computer program, when executed by the processor, causes the processor to perform the following operations: inputting an annotated training sample into an abstract description model; training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment; obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network; obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0107] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconfiguration network.
[0108] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into an LSTM hidden layer of the current moment in the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network; and inputting the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0109] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0110] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting an annotated training sample into the encoding network, and extracting a feature of training information to obtain an eigenvector of the training information; inputting the eigenvector into the decoding network to obtain a hidden state of each moment; generating a word corresponding to a current moment according to the hidden state of each moment; obtaining a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment; and obtaining a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0111] In an embodiment, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. The computer program, when executed by a processor, causes the processor to perform the following operations: obtaining input information; and inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0112] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting an annotated training sample into the abstract description model; training the encoding network and the decoding network of the abstract description model based on supervision of the first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment; obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network; obtaining a value of the second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0113] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconfiguration network.
[0114] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into an LSTM hidden layer of the current moment of the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network; and inputting the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0115] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0116] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting an annotated training sample into the encoding network, and extracting a feature of training information to obtain an eigenvector of the training information; inputting the eigenvector into the decoding network to obtain a hidden state of each moment; generating a word corresponding to a current moment according to the hidden state of each moment; obtaining a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment; and obtaining a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0117] In an embodiment, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. The computer program, when executed by a processor, causes the processor to perform the following operations: inputting an annotated training sample into an abstract description model; training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using an RNN to obtain a hidden state of each current moment; obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network; obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0118] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconfiguration network.
[0119] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into an LSTM hidden layer of the current moment of the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network; and inputting the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0120] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0121] In an embodiment, the computer program, when executed by the processor, causes the processor to perform the following operations: inputting an annotated training sample into the encoding network, and extracting a feature of training information to obtain an eigenvector of the training information; inputting the eigenvector into the decoding network to obtain a hidden state of each moment; generating a word corresponding to a current moment according to the hidden state of each moment; obtaining a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment; and obtaining a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0122] A person of ordinary skill in the art may understand that all or some of the processes in the foregoing embodiments may be implemented by a computer program instructing relevant hardware. The computer program may be stored in a non-volatile computer-readable storage medium. When the program is executed, the processes of the embodiments of the foregoing methods may be included. Any reference to the processor, storage, database, or another medium that is used in the embodiments provided by this application may include a non-volatile and/or volatile memory. The non-volatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or flash memory. The non-volatile memory may include a random access memory (RAM) or an external cache. By way of description and not limitation, the RAM is available in a plurality of forms, for example, a static RAM (SRAM), a dynamic RAM (DRAM), a synchronous DRAM (SDRAM), a double data rate SDRAM (DDRSDRAM), an enhanced SDRAM (ESDRAM), a synchlink DRAM (SLDRAM), a Rambus direct RAM (RDRAM), a direct Rambus dynamic RAM (DRDRAM), a Rambus dynamic RAM (RDRAM), and the like.
[0123] Technical features of the foregoing embodiments may be combined randomly, and for brevity of description, not all possible combinations of the technical features in the foregoing embodiments are described. However, as long as there is no contradiction in the combinations of these technical features, the combinations shall be considered as falling within the scope recorded in this specification.
[0124] The foregoing embodiments only describe several implementations of this application, and the description thereof is specific and detailed, but cannot therefore be understood as a limitation to the patent scope of this application. A person of ordinary skill in the art may further make variations and improvements without departing from the conception of this application, and these all fall within the protection scope of this application. Therefore, the patent protection scope of this application shall be subject to the appended claims.

权利要求:
Claims (20)
[0001] An abstract description generating method, implemented by a computer device, the method comprising:
obtaining input information; and
inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0002] The method according to claim 1, further comprising:
inputting an annotated training sample into the abstract description model;
training the encoding network and the decoding network of the abstract description model based on the supervision of the first loss function, wherein the encoding network obtains an eigenvector of the training sample, and the decoding network decodes the eigenvector by using a recurrent neural network (RNN) to obtain a hidden state of each current moment;
obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network;
obtaining a value of the second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and
obtaining a final parameter of the abstract description model determined based on the supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0003] The method according to claim 2, wherein the operation of obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network comprises: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment in a reconfiguration network.
[0004] The method according to claim 3, wherein the operation of obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment in a reconfiguration network comprises:
inputting the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into a long short-term memory (LSTM) hidden layer of the current moment in the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network; and
inputting the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0005] The method according to claim 2, wherein the operation of obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network comprises: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0006] The method according to claim 2, wherein the operation of training the encoding network and the decoding network of the abstract description model based on the supervision of the first loss function comprises:
inputting the annotated training sample into the encoding network, and extracting a feature of training information, to obtain an eigenvector of the training information;
inputting the eigenvector into the decoding network to obtain a hidden state of each moment;
generating a word corresponding to a current moment according to the hidden state of each moment;
obtaining a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment; and
obtaining a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0007] An abstract description model training method, implemented by a computer device, the method comprising:
inputting an annotated training sample into an abstract description model;
training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using a recurrent neural network (RNN) to obtain a hidden state of each current moment;
obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network;
obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and
obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0008] The method according to claim 7, wherein the operation of obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network comprises: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment in a reconfiguration network.
[0009] The method according to claim 8, wherein the operation of obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment in a reconfiguration network comprises:
inputting the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into a long short-term memory (LSTM) hidden layer of the current moment in the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network; and
inputting the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0010] The method according to claim 7, wherein the operation of obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network comprises: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0011] The method according to claim 7, wherein the operation of training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function comprises:
inputting the annotated training sample into the encoding network, and extracting a feature of the training information to obtain an eigenvector of the training information;
inputting the eigenvector into the decoding network to obtain a hidden state of each moment;
generating a word corresponding to a current moment according to the hidden state of each moment;
obtaining a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment; and
obtaining a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0012] A computer device, comprising a memory and a processor, the memory storing a computer program, and the computer program, when executed by the processor, causing the processor to perform the following operations:
obtaining input information; and
inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0013] A computer device, comprising: a memory and a processor, the memory storing a computer program, and the computer program, when executed by the processor, causing the processor to perform the following operations:
inputting an annotated training sample into an abstract description model;
training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of a training sample, and the decoding network decoding the eigenvector by using a recurrent neural network (RNN) to obtain a hidden state of each current moment;
obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network;
obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and
obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0014] The computer device according to claim 13, wherein the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconstruction network.
[0015] The computer device according to claim 14, wherein the computer program, when executed by the processor, causes the processor to perform the following operations:
inputting the hidden state of each current moment outputted by the decoding network and a hidden state of the previous moment in the reconfiguration network into a long short-term memory (LSTM) hidden layer of each current moment in the reconfiguration network, to obtain a hidden state of the current moment in the reconfiguration network; and
inputting the hidden state of the current moment in the reconfiguration network into a fully connected layer, to obtain the backward-derived hidden state of the previous moment.
[0016] The computer device according to claim 13, wherein the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining, by using a back propagation algorithm, the backward-derived hidden state of the previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network.
[0017] The computer device according to claim 13, wherein the computer program, when executed by the processor, causes the processor to perform the following operations:
inputting the annotated training sample into the encoding network, and extracting a feature of the training information to obtain an eigenvector of training information;
inputting the eigenvector into the decoding network to obtain a hidden state of each moment;
generating a word corresponding to a current moment according to the hidden state of each moment;
obtaining a value of the first loss function according to the generated word corresponding to the current moment and an actual annotated word of the current moment; and
obtaining a preliminary parameter of the abstract description model in a case that the value of the first loss function reaches a preset value.
[0018] A computer-readable storage medium, the computer-readable storage medium storing a computer program, and the computer program, when executed by a processor, causing the processor to perform the following operations:
obtaining input information; and
inputting the input information into a pre-trained abstract description model, obtaining an eigenvector of the input information through an encoding network of the abstract description model, and decoding the eigenvector through a decoding network of the abstract description model to generate abstract description of the input information, the encoding network and the decoding network being pre-trained based on supervision of a first loss function, a backward-derived hidden state of a previous moment being obtained through backward derivation according to a hidden state of each moment outputted by the encoding network, and the abstract description model determined based on supervision of a second loss function being obtained according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network.
[0019] A computer-readable storage medium, the computer-readable storage medium storing a computer program, and the computer program, when executed by a processor, causing the processor to perform the following operations:
inputting an annotated training sample into an abstract description model;
training an encoding network and a decoding network of the abstract description model based on supervision of a first loss function, the encoding network obtaining an eigenvector of the training sample, and the decoding network decoding the eigenvector by using a recurrent neural network (RNN) to obtain a hidden state of each current moment;
obtaining a backward-derived hidden state of a previous moment through backward derivation according to the hidden state of each moment outputted by the decoding network;
obtaining a value of a second loss function according to the backward-derived hidden state of the previous moment and an actual hidden state of the previous moment outputted by the decoding network; and
obtaining a final parameter of the abstract description model determined based on supervision of the second loss function in a case that the value of the second loss function reaches a preset value.
[0020] The computer-readable storage medium according to claim 19, wherein the computer program, when executed by the processor, causes the processor to perform the following operation: obtaining the backward-derived hidden state of the previous moment through backward derivation by using the hidden state of each moment outputted by the decoding network as input of each corresponding moment of a reconstruction network.

类似技术:

---

公开号 | 公开日 | 专利标题

---

US10922488B1|2021-02-16|Computing numeric representations of words in a high-dimensional space

---

Srivastava et al.2017|Veegan: Reducing mode collapse in gans using implicit variational learning

---

Maddox et al.2019|A simple baseline for bayesian uncertainty in deep learning

---

US10621420B2|2020-04-14|Generating numeric embeddings of images

---

Xiao et al.2017|Attentional factorization machines: Learning the weight of feature interactions via attention networks

---

Yang et al.2017|Breaking the softmax bottleneck: A high-rank RNN language model

---

Zhao et al.2015|Bayesian CP factorization of incomplete tensors with automatic rank determination

---

Zhao et al.2015|Bayesian robust tensor factorization for incomplete multiway data

---

US10417562B2|2019-09-17|Batch normalization layers

---

US10762422B2|2020-09-01|Wide and deep machine learning models

---

Hensman et al.2013|Gaussian processes for big data

---

Rodriguez et al.2013|Fast principal component pursuit via alternating minimization

---

US9807473B2|2017-10-31|Jointly modeling embedding and translation to bridge video and language

---

US20180124331A1|2018-05-03|Video retrieval system using adaptive spatiotemporal convolution feature representation with dynamic abstraction for video to language translation

---

Sedghi et al.2014|Provable methods for training neural networks with sparse connectivity

---

Ververidis et al.2008|Fast and accurate sequential floating forward feature selection with the Bayes classifier applied to speech emotion recognition

---

Hsieh et al.2015|PU learning for matrix completion

---

US10977547B2|2021-04-13|Convolutional gated recurrent neural networks

---

US20200134470A1|2020-04-30|Compressed recurrent neural network models

---

Bucak et al.2009|Incremental subspace learning via non-negative matrix factorization

---

Park et al.2013|Hierarchical bayesian matrix factorization with side information

---

US10248664B1|2019-04-02|Zero-shot sketch-based image retrieval techniques using neural networks for sketch-image recognition and retrieval

---

Zhao et al.2018|Unsupervised discrete sentence representation learning for interpretable neural dialog generation

---

US20170372487A1|2017-12-28|Eye gaze tracking using neural networks

---

US20200411032A1|2020-12-31|Generating audio using neural networks

同族专利:

---

公开号 | 公开日

---

CN108334889B|2020-04-03|

---

WO2019105157A1|2019-06-06|

---

CN110598779A|2019-12-20|

---

CN108334889A|2018-07-27|

---

US20200082271A1|2020-03-12|

引用文献:

---

公开号 | 申请日 | 公开日 | 申请人 | 专利标题

---

法律状态:
2019-06-08| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |

---

2020-06-19| PUAI| Public reference made under article 153(3) epc to a published international application that has entered the european phase|Free format text: ORIGINAL CODE: 0009012 |

---

2020-06-19| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |

---

2020-07-22| 17P| Request for examination filed|Effective date: 20191120 |

---

2020-07-22| AK| Designated contracting states|Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |

---

2020-07-22| AX| Request for extension of the european patent|Extension state: BA ME |

---

2021-03-10| DAV| Request for validation of the european patent (deleted)|

---

2021-03-10| DAX| Request for extension of the european patent (deleted)|

优先权:

---

申请号 | 申请日 | 专利标题

---